Does Eye Contact Still Matter in an AI Video Interview? What Two New Studies Found

The 2024 Study: Genuine Camera Gaze Measurably Changes How You're Scored

Published in Scientific Reports in May 2024 — a peer-reviewed, Q1-ranked journal — a study out of Hiroshima University set up as close to a controlled experiment on this exact question as this research has gotten. Twelve interviewees were each recorded twice, giving the same simulated interview responses once while looking directly at the camera and once while looking down at a screen below it, removing content and delivery as variables since the same person gave the same answer both times.

Thirty-eight full-time working adults then evaluated the recordings — plus a third, video-free, audio-only condition — across six criteria: intimacy, social desirability, job abilities, decisiveness, cooperativeness, and overall hireability.

The direct-gaze recordings scored significantly higher than the downward-gaze recordings across these measures. More strikingly, the downward-gaze condition scored even worse than the audio-only condition with no video at all — visibly looking away from the camera cost candidates more than having no camera in the picture whatsoever.

One more finding worth knowing: female evaluators judged downward gaze more harshly than male evaluators did, with the evaluation gap widest when a female interviewee looked away. The study's own authors flagged this as a sign the effect could interact with real-world demographic bias in ways a lab setting wouldn't fully capture — a caution about the finding's edges, not a reason to dismiss the core result.

“Looking away from the camera scored worse than giving the interview with no camera at all.”

The 2025 Twist: Faking Eye Contact Doesn't Work the Way People Assume

A second, newer study complicates the obvious takeaway. Presented at ACM CHI 2025 in April 2025, researchers at Virginia Tech (Jelson, Tausif, Lim, Khanna, and Lee) ran two linked studies on a related but different question — not whether real eye contact matters, but whether simulated eye contact, the kind produced by gaze-correction software that digitally redirects where your eyes appear to look, has the same effect.

The first study had participants act as interviewers evaluating a confederate interviewee whose eye contact was simulated for half the interview. A second, larger online survey had experienced interviewers independently rate interview videos, half featuring the same simulated gaze, to check whether the first result held at scale. In both, simulated eye contact had little measurable effect on evaluation scores — far less than the common assumption that correcting your gaze should help. The researchers went further, raising a specific concern: if evaluators noticed the correction, the effect could run the other way, reading as inauthentic or manipulated rather than confident.

Put the two studies side by side and the finding isn't contradictory — it's specific. The 2024 Hiroshima study tested genuine behavior: did this real person actually look at the camera or not. The 2025 Virginia Tech study tested a technical substitute for that behavior: can software make it look like you did, without you actually doing it. Only the real behavior showed a clear benefit.

“Simulated eye contact had little measurable effect — far less than common belief would predict.”

Is an Algorithm Actually Grading Your Eyes, Though?

Neither study tested an AI interview-scoring algorithm. Both tested human evaluators — full-time workers in the 2024 study, experienced interviewers in the 2025 one — rating recorded video, not a company's proprietary hiring model. That distinction matters for what these findings do and don't prove.

On what today's major AI interview platforms actually score, this blog has covered the shift away from facial analysis in detail: HireVue, the largest vendor in the category, publicly dropped facial and gaze analysis from its algorithmic scoring in January 2021, after an external audit and sustained criticism that the practice was scientifically shaky and prone to encoding invisible bias.

That same shift is why most one-way, no-interviewer AI video interviews now evaluate your transcript rather than your face — the algorithmic first pass is built to read what you said, not where you looked.

So the honest, two-part picture: there's no confirmed evidence that a mainstream AI scoring algorithm independently grades your gaze in 2026. But almost every AI-moderated interview format still routes to a human somewhere in its pipeline — the recruiter reviewing a shortlist after the algorithm's first pass, or the hiring panel on a live video call.

That's the same human whose snap first impression research on recruiter attention has already shown forms faster than most candidates assume, and the 2024 gaze study measured exactly that reaction.

“The algorithm may not be watching your eyes. The person deciding whether to move you forward still might be.”

What This Means for How You Sit in Front of the Camera

Three practical conclusions follow directly from what the research actually found, rather than from the louder, vaguer fear that usually drives interview-prep advice.

• Look at the lens yourself, especially in the first several seconds of an answer — the 2024 research suggests this is where a human reviewer's impression forms, and it's the one lever with a measured, positive effect.
• Skip gaze-correction software or filters for this specific purpose — the 2025 research found it does little for your score and risks reading as manipulated if an evaluator notices it, a worse outcome than imperfect, genuine eye contact.
• Don't spend prep time worrying that an algorithm is logging every glance at your notes — put that energy into answer structure instead, which is what the scoring systems that actually run first are built to evaluate.

That last point is the bigger lever by a wide margin: a clear, specific, complete answer reads the same whether a transcript-based algorithm or a tired human reviewer is doing the reading, and it's worth more of your prep time than camera position ever will be.

The first few seconds carry disproportionate weight before you've said anything substantive at all — the same timing effect shows up well before an AI-moderated interview even starts, one more reason genuine camera position is worth settling before you hit record, not mid-answer.

A low-tech trick that solves most of the genuine-gaze problem without any software at all: tape a small arrow or sticky note just above your webcam lens, and keep any notes directly beneath it rather than off to the side — close enough that a natural downward glance barely shifts your visible gaze.

Key Takeaways

If you remember only a few things, remember these:

• A 2024 Scientific Reports study found genuine on-camera gaze scored significantly better than looking away — and looking away scored worse than having no video at all.
• A 2025 Virginia Tech/ACM CHI study found simulated, software-corrected eye contact had little measurable effect, and risks reading as inauthentic if noticed.
• Neither study tested an actual AI scoring algorithm — both measured human evaluators. HireVue, the largest AI interview vendor, has not used facial or gaze analysis in its algorithmic scoring since January 2021.
• A human reviewer is still likely somewhere in the pipeline, so genuine camera eye contact is still worth the effort — gaze-correction software is not.
• Answer structure and content matter more than camera position to the scoring systems that run first; don't let eye-contact anxiety crowd out the prep time that belongs there.